Privacy-Preserving Federated Spatiotemporal Dynamic Graph Neural Network Framework for Epileptic Seizure Prediction

This paper proposes a federated learning framework for epilepsy seizure prediction that addresses privacy concerns and data-sharing constraints in clinical environments. The approach uses dynamic graph neural networks to model the time-varying topological characteristics of EEG signals, improving upon traditional static modeling methods that fail to capture high-order nonlinear correlations in brain functional networks.